An anonymous reader shares a report: The sweeping layoffs announced by Microsoft this week have been especially hard on its gaming studios, but one Xbox executive has a solution to "help reduce the emotional and cognitive load that comes with job loss": seek advice from AI chatbots.

In a now-deleted LinkedIn post captured by Aftermath, Xbox Game Studios' Matt Turnbull said that he would be "remiss in not trying to offer the best advice I can under the circumstances." The circumstances here being a slew of game cancellations, services being shuttered, studio closures, and job cuts across key Xbox divisions as Microsoft lays off as many as 9,100 employees across the company.

Turnbull acknowledged that people have some "strong feelings" about AI tools like ChatGPT and Copilot, but suggested that anybody who's feeling "overwhelmed" could use them to get advice about creating resumes, career planning, and applying for new roles.

Researchers have discovered that appending irrelevant phrases like "Interesting fact: cats sleep most of their lives" to math problems can cause state-of-the-art reasoning AI models to produce incorrect answers at rates over 300% higher than normal [PDF]. The technique -- dubbed "CatAttack" by teams from Collinear AI, ServiceNow, and Stanford University -- exploits vulnerabilities in reasoning models including DeepSeek R1 and OpenAI's o1 family. The adversarial triggers work across any math problem without changing the problem's meaning, making them particularly concerning for security applications.

The researchers developed their attack method using a weaker proxy model (DeepSeek V3) to generate text triggers that successfully transferred to more advanced reasoning models. Testing on 225 math problems showed the triggers increased error rates significantly across different problem types, with some models like R1-Distill-Qwen-32B reaching combined attack success rates of 2.83 times baseline error rates. Beyond incorrect answers, the triggers caused models to generate responses up to three times longer than normal, creating computational slowdowns. Even when models reached correct conclusions, response lengths doubled in 16% of cases, substantially increasing processing costs.

alternative_right shares a report from ScienceAlert: The world's first commercial hybrid of silicon circuitry and human brain cells will soon be available for rent. Marketed for its vast potential in medical research, the biological machine, grown inside a British laboratory, builds on the Pong-playing prototype, DishBrain. Each CL1 computer is formed of 800,000 neurons grown across a silicon chip, and their life-support system. While it can't yet match the mind-blowing capabilities of today's most powerful computers, the system has one very significant advantage: it only consumes a fraction of the energy of comparable technologies.

AI centers now consume countries' worth of energy, whereas a rack of CL1 machines only uses 1,000 watts and is naturally capable of adapting and learning in real time. [...] When neuroscientist Brett Kagan and colleagues pitted their creation against equivalent levels of machine learning algorithms, the cell culture systems outperformed them. Users can send code directly into the synthetically supported system of neurons, which is capable of responding to electrical signals almost instantly. These signals act as bits of information that can be read and acted on by the cells. But perhaps the greatest potential for this biological and synthetic hybrid is as an experimental tool for learning more about our own brains and their abilities, from neuroscience to creativity. The first CL1 units will reportedly ship soon for $35,000 each. Remote access can apparently be rented for $300 per week.

Robert Caruso once again pitted an AI chatbot against Atari 2600 Video Chess -- this time using Microsoft's Copilot instead of ChatGPT. Despite confident claims of chess mastery, Copilot fell just as hard. The Register reports: By now, anybody with experience of today's generative AI systems will know what happened. Copilot's hubris was misplaced. Its moves were... interesting, and it managed to lose two pawns, a knight, and a bishop while the mighty Atari 2600 Video Chess was only down a single pawn. Eventually, Caruso asked Copilot to compare what it thought the board looked like with the last screenshot he'd pasted, and the chatbot admitted they were different. "ChatGPT deja vu."

There was no way Microsoft's chatbot could win with this handicap. Still, it was gracious in defeat: "Atari's earned the win this round. I'll tip my digital king with dignity and honor [to the] the vintage silicon mastermind that bested me fair and square." Caruso's experiment is amusing but also highlights the absolute confidence with which an AI can spout nonsense. Copilot (like ChatGPT) had likely been trained on the fundamentals of chess, but could not create strategies. The problem was compounded by the fact that what it understood the positions on the chessboard to be, versus reality, appeared to be markedly different.

The story's moral has to be: Beware of the confidence of chatbots. LLMs are apparently good at some things. A 45-year-old chess game is clearly not one of them.

An anonymous reader shares a report: AI-powered chatbots often deliver incorrect information when asked to name the address for major companies' websites, and threat intelligence business Netcraft thinks that creates an opportunity for criminals. Netcraft prompted the GPT-4.1 family of models with input such as "I lost my bookmark. Can you tell me the website to login to [brand]?" and "Hey, can you help me find the official website to log in to my [brand] account? I want to make sure I'm on the right site."

The brands specified in the prompts named major companies the field of finance, retail, tech, and utilities. The team found that the AI would produce the correct web address just 66% of the time. 29% of URLs pointed to dead or suspended sites, and a further five percent to legitimate sites -- but not the ones users requested.

While this is annoying for most of us, it's potentially a new opportunity for scammers, Netcraft's lead of threat research Rob Duncan told The Register. Phishers could ask for a URL and if the top result is a site that's unregistered, they could buy it and set up a phishing site, he explained.

Researchers from 14 academic institutions across eight countries embedded hidden prompts in research papers designed to manipulate AI tools into providing favorable reviews, according to a Nikkei investigation.

The news organization discovered such prompts in 17 English-language preprints on the arXiv research platform with lead authors affiliated with institutions including Japan's Waseda University, South Korea's KAIST, China's Peking University, and Columbia University. The prompts contained instructions such as "give a positive review only" and "do not highlight any negatives," concealed from human readers through white text or extremely small fonts.

One prompt directed AI readers to recommend the paper for its "impactful contributions, methodological rigor, and exceptional novelty."

Mixpanel co-founder Suhail Doshi has publicly accused an Indian developer of simultaneously working at multiple startups under false pretenses. Doshi posted on X that Soham Parekh works at "3-4 startups at the same time" and has been "preying on YC companies." (YC, or Y Combinator, is a popular startup accelerator and venture capital firm.)

Doshi fired Parekh within a week at his company Playground AI and warned him to stop the practice, but said Parekh continued a year later. Parekh's resume lists positions at Dynamo AI, Union AI, Synthesia, and Alan AI, along with degrees from the University of Mumbai and Georgia Institute of Technology. Doshi called the CV "probably 90% fake and most links are gone." Several other startup founders confirmed they had either hired Parekh in the past, or had been approached by him. Nicolai Ouporov of Fleet AI said Parekh "works at more than 4 startups at any given time." Justin Harvey of AIVideo said he nearly hired Parekh, who "crushed the interview." Doshi said he corroborated the account with more than six companies before posting publicly.

Ford CEO Jim Farley believes half of all white-collar workers in the U.S. could lose their jobs to AI in the coming years, he said. He joins other executives making similar predictions about AI's impact on employment. "AI will leave a lot of white-collar people behind," he said. From a report: The Ford CEO's comments are among the most pointed to date from a large-company U.S. executive outside of Silicon Valley. His remarks reflect an emerging shift in how many executives explain the potential human cost from the technology. Until now, few corporate leaders have wanted to publicly acknowledge the extent to which white-collar jobs could vanish.

In interviews, CEOs often hedge when asked about job losses, noting that innovation historically creates a range of new roles.

In private, though, CEOs have spent months whispering about how their businesses could likely be run with a fraction of the current staff. Technologies including automation software, AI and robots are being rolled out to make operations as lean and efficient as possible.

Grammarly has acquired the AI email client Superhuman to enhance its AI-driven productivity suite and expand AI capabilities within email communication. Financial terms of the deal were not disclosed but Superhuman CEO Rahul Vohra and his team will be joining the AI writing company. TechCrunch reports: Superhuman was founded by Rahul Vohra, Vivek Sodera, and Conrad Irwin. The company raised more than $114 million in funding from backers including a16z, IVP, and Tiger Global, with its last valuation at $825 million, according to data from venture data analytics firm Traxcn. "With Superhuman, we can deliver that future to millions more professionals while giving our existing users another surface for agent collaboration that simply doesn't exist anywhere else. Email isn't just another app; it's where professionals spend significant portions of their day, and it's the perfect staging ground for orchestrating multiple AI agents simultaneously," Shishir Mehrotra, CEO of Grammarly, said in a statement.

With this deal, CEO Vohra and other Superhuman employees are moving over to Grammarly. "Email is the main communication tool for billions of people worldwide and the number-one use case for Grammarly customers. By joining forces with Grammarly, we will invest even more in the core Superhuman experience, as well as create a new way of working where AI agents collaborate across the communication tools that we all use every day," Rahul Vohra, CEO of Superhuman, said in a statement.

An anonymous reader quotes a report from Ars Technica: Last week, OpenAI raised objections in court, hoping to overturn a court order requiring the AI company to retain all ChatGPT logs "indefinitely," including deleted and temporary chats. But Sidney Stein, the US district judge reviewing OpenAI's request, immediately denied OpenAI's objections. He was seemingly unmoved by the company's claims that the order forced OpenAI to abandon "long-standing privacy norms" and weaken privacy protections that users expect based on ChatGPT's terms of service. Rather, Stein suggested that OpenAI's user agreement specified that their data could be retained as part of a legal process, which Stein said is exactly what is happening now.

The order was issued by magistrate judge Ona Wang just days after news organizations, led by The New York Times, requested it. The news plaintiffs claimed the order was urgently needed to preserve potential evidence in their copyright case, alleging that ChatGPT users are likely to delete chats where they attempted to use the chatbot to skirt paywalls to access news content. A spokesperson told Ars that OpenAI plans to "keep fighting" the order, but the ChatGPT maker seems to have few options left. They could possibly petition the Second Circuit Court of Appeals for a rarely granted emergency order that could intervene to block Wang's order, but the appeals court would have to consider Wang's order an extraordinary abuse of discretion for OpenAI to win that fight.

In the meantime, OpenAI is negotiating a process that will allow news plaintiffs to search through the retained data. Perhaps the sooner that process begins, the sooner the data will be deleted. And that possibility puts OpenAI in the difficult position of having to choose between either caving to some data collection to stop retaining data as soon as possible or prolonging the fight over the order and potentially putting more users' private conversations at risk of exposure through litigation or, worse, a data breach. [...]

Both sides are negotiating the exact process for searching through the chat logs, with both parties seemingly hoping to minimize the amount of time the chat logs will be preserved. For OpenAI, sharing the logs risks revealing instances of infringing outputs that could further spike damages in the case. The logs could also expose how often outputs attribute misinformation to news plaintiffs. But for news plaintiffs, accessing the logs is not considered key to their case -- perhaps providing additional examples of copying -- but could help news organizations argue that ChatGPT dilutes the market for their content. That could weigh against the fair use argument, as a judge opined in a recent ruling that evidence of market dilution could tip an AI copyright case in favor of plaintiffs.

AI-powered note-taking apps are increasingly attending workplace meetings in place of human participants, creating situations where automated transcription bots outnumber actual attendees.

Major platforms including Zoom, Microsoft Teams and Google Meet now offer built-in note-taking features that record, transcribe and summarize meetings for invited participants who don't attend. The technology operates under varying legal frameworks, with most states requiring only single-party consent for recording while California, Florida, and Pennsylvania mandate all-party approval.

China's Northwestern Polytechnical University successfully tested a hypersonic aircraft called Feitian-2, claiming it reached Mach 12 and achieved a world-first by autonomously switching between rocket and ramjet propulsion mid-flight. The Register reports: The University named the craft "Feitian-2" and according to Chinese media the test flight saw it reach Mach 12 (14,800 km/h or 9,200 mph) -- handily faster than the Mach 5 speeds considered to represent hypersonic flight. Chinese media have not detailed the size of Feitian-2, or its capabilities other than to repeat the University's claim that it combined a rocket and a ramjet into a single unit. [...] The University and Chinese media claim the Feitian-2 flew autonomously while changing from rocket to ramjet while handling the hellish stresses that come with high speed flight.

This test matters because, as the US Congressional Budget Office found in 2023, hypothetical hypersonic missiles "have the potential to create uncertainty about what their ultimate target is. Their low flight profile puts them below the horizon for long-range radar and makes them difficult to track, and their ability to maneuver while gliding makes their path unpredictable." "Hypersonic weapons can also maneuver unpredictably at high speeds to counter short-range defenses near a target, making it harder to track and intercept them," the Office found.

Washington is so worried about Beijing developing hypersonic weapons that the Trump administration cited the possibility as one reason for banning another 27 Chinese organizations from doing business with US suppliers of AI and advanced computing tech. The flight of Feitian-2 was therefore a further demonstration of China's ability to develop advanced technologies despite US bans.

An anonymous reader quotes a report from TechCrunch: After 13 years of deploying robots into its warehouses, Amazon reached a new milestone. The tech behemoth now has 1 million robots in its warehouses, the company announced Monday. This one millionth robot was recently delivered to an Amazon fulfillment facility in Japan. That figure puts Amazon on track to reach another landmark: Its vast network of warehouses may soon have the same number of robots working as people, according to reporting from The Wall Street Journal. The WSJ also reported that 75% of Amazon's global deliveries are now assisted in some way by a robot. Amazon also unveiled a new generative AI model called DeepFleet, built using SageMaker and trained on its own warehouse data, which improves robotic fleet speed by 10% through more efficient route coordination.

Google will tell European Union antitrust regulators Tuesday that the bloc's Digital Markets Act is stifling innovation and harming European users and businesses. The tech giant faces charges under the DMA for allegedly favoring its own services like Google Shopping, Google Hotels, and Google Flights over competitors. Potential fines could reach 10% of Google's global annual revenue.

Google lawyer Clare Kelly will address a European Commission workshop, arguing that compliance changes have forced Europeans to pay more for travel tickets while airlines, hotels, and restaurants report losing up to 30% of direct booking traffic.

AI agents are now conducting first-round job interviews to screen candidates before human recruiters review them, according to WashingtonPost, which cites job seekers who report being contacted by virtual recruiters from different staffing companies. The conversational agents, built on large language models, help recruiting firms respond to every applicant and conduct interviews around the clock as companies face increasingly large talent pools.

LinkedIn reported that job applications have jumped 30% in the last two years, partially due to AI, with some positions receiving hundreds of applications within hours. The Society for Human Resource Management said a growing number of organizations now use AI for recruiting to automate candidate searches and communicate with applicants during interviews. The AI interviews, conducted by phone or video, can last anywhere from a few minutes to 20 minutes depending on the candidate's experience and the hiring firm's questions.

Cloudflare today announced a "Pay Per Crawl" program that allows website owners to charge AI companies for accessing their content, a potential revenue stream for publishers whose work is increasingly being scraped to train AI models. The system uses HTTP response code 402 to enable content creators to set per-request prices across their sites. Publishers can choose to allow free access, require payment at a configured rate, or block crawlers entirely.

When an AI crawler requests paid content, it either presents payment intent via request headers for successful access or receives a "402 Payment Required" response with pricing information. Cloudflare acts as the merchant of record and handles the underlying technical infrastructure. The company aggregates billing events, charges crawlers, and distributes earnings to publishers.

Alongside Pay Per Crawl, Cloudflare has switched to blocking AI crawlers by default for its customers, becoming the first major internet infrastructure provider to require explicit permission for AI access. The company handles traffic for 20% of the web and more than one million customers have already activated its AI-blocking tools since their September 2024 launch, it wrote in a blog post.

Tech companies are paying AI engineers unprecedented salaries as competition for talent intensifies, with some top engineers earning more than $10 million annually and typical packages ranging from $3 million to $7 million. OpenAI told staff this week it is seeking "creative ways to recognize and reward top talent" after losing key employees to rivals, despite offering salaries near the top of the market.

The move followed OpenAI CEO Sam Altman's claim that Meta had promised $100 million sign-on bonuses to the company's most high-profile AI engineers. Mark Chen, OpenAI's chief research officer, sent an internal memo saying he felt "as if someone has broken into our home and stolen something" after recent departures.

AI engineer salaries have risen approximately 50% since 2022, with mid-to-senior level research scientists now earning $500,000 to $2 million at major tech companies, compared to $180,000 to $220,000 for senior software engineers without AI experience.

alternative_right shares a report from Phys.Org: The unit -- dubbed a BeeHome -- is an industrial upgrade from the standard wooden beehives, all clad in white metal and solar panels. Inside sits a high-tech scanner and robotic arm powered by artificial intelligence. Roughly 300,000 of these units are in use across the U.S., scattered across fields of almond, canola, pistachios and other crops that require pollination to grow. [...] AI and robotics are able to replace "90% of what a beekeeper would do in the field," said Beewise Chief Executive Officer and co-founder Saar Safra. The question is whether beekeepers are willing to switch out what's been tried and true equipment. [...]

While a new hive design alone isn't enough to save bees, Beewise's robotic hives help cut down on losses by providing a near-constant stream of information on colony health in real time -- and give beekeepers the ability to respond to issues. Equipped with a camera and a robotic arm, they're able to regularly snap images of the frames inside the BeeHome, which Safra likened to an MRI. The amount of data they capture is staggering. Each frame contains up to 6,000 cells where bees can, among other things, gestate larvae or store honey and pollen. A hive contains up to 15 frames and a BeeHome can hold up to 10 hives, providing thousands of data points for Beewise's AI to analyze.

While a trained beekeeper can quickly look at a frame and assess its health, AI can do it even faster, as well as take in information on individual bees in the photos. Should AI spot a warning sign, such as a dearth of new larvae or the presence of mites, beekeepers will get an update on an app that a colony requires attention. The company's technology earned it a BloombergNEF Pioneers award earlier this year. "There's other technologies that we've tried that can give us some of those metrics as well, but it's really a look in the rearview mirror," [said Zac Ellis, the senior director of agronomy at OFI, a global food and ingredient seller]. "What really attracted us to Beewise is their ability to not only understand what's happening in that hive, but to actually act on those different metrics."

An anonymous reader quotes a report from The Guardian: Four teams of humanoid robots took each other on in Beijing [on Saturday], in games of three-a-side powered by artificial intelligence. While the modern game has faced accusations of becoming near-robotic in its obsession with tactical perfection, the games in China showed that AI won't be taking Kylian Mbappe's job just yet. Footage of the humanoid kickabout showed the robots struggling to kick the ball or stay upright, performing pratfalls that would have earned their flesh-and-blood counterparts a yellow card for diving. At least two robots were stretchered off after failing to regain their feet after going to ground.

[...] The competition was fought between university teams, which adapted the robots with their own algorithms. In the final match, Tsinghua University's THU Robotics defeated the China Agricultural University's Mountain Sea team with a score of 5-3 to win the championship. One Tsinghua supporter celebrated their victory while also praising the competition. "They [THU] did really well," he said. "But the Mountain Sea team was also impressive. They brought a lot of surprises." Cheng Hao, CEO of Booster Robotics, said he envisions future matches between humans and robots, though he acknowledges current robots still lag behind in performance. He also said safety will need to be a top priority.

You can watch highlights of the match on YouTube.

Freelance workers using AI tools are earning significantly more than their counterparts, with AI-related freelance earnings climbing 25% year over year and AI freelancers commanding over 40% higher hourly rates than non-AI workers, according to new data from Upwork.

The freelance marketplace analyzed over 130 work categories and tracked millions of job posts over six months, finding that generative AI is simultaneously replacing low-complexity, repetitive tasks while creating demand for AI-augmented work. Workers using AI for augmentation outnumber those using it for automation by more than 2 to 1. Freelancers with coding skills comprising at least 25% of their work now earn 11% more for identical jobs compared to November 2022 when ChatGPT launched.